-
Notifications
You must be signed in to change notification settings - Fork 11
/
opts.py
274 lines (271 loc) · 12.8 KB
/
opts.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
import argparse
from pathlib import Path
def parse_opts():
parser = argparse.ArgumentParser()
parser.add_argument('--root_path',
default=None,
type=Path,
help='Root directory path')
parser.add_argument('--video_path',
default=None,
type=Path,
help='Directory path of videos')
parser.add_argument('--audio_path',
default=None,
type=Path,
help='Directory path of audios')
parser.add_argument('--annotation_path',
default=None,
type=Path,
help='Annotation file path')
parser.add_argument('--result_path',
default=None,
type=Path,
help='Result directory path')
parser.add_argument('--dataset',
default='kinetics',
type=str,
help='Used dataset (activitynet | kinetics | ucf101 | hmdb51)')
parser.add_argument('--n_classes',
default=400,
type=int,
help=
'Number of classes (activitynet: 200, kinetics: 400 or 600, ucf101: 101, hmdb51: 51)')
parser.add_argument('--n_pretrain_classes',
default=0,
type=int,
help=('Number of classes of pretraining task.'
'When using --pretrain_path, this must be set.'))
parser.add_argument('--pretrain_path',
default=None,
type=Path,
help='Pretrained model path (.pth).')
parser.add_argument('--ft_begin_module',
default='',
type=str,
help=('Module name of beginning of fine-tuning'
'(conv1, layer1, fc, denseblock1, classifier, ...).'
'The default means all layers are fine-tuned.'))
parser.add_argument('--sample_size',
default=112,
type=int,
help='Height and width of inputs')
parser.add_argument('--image_size',
default=112,
type=int,
help='Height and width of input images, '
'higher input size (e.g. 224) can boost performance')
parser.add_argument('--sample_duration',
default=16,
type=int,
help='Temporal duration of inputs')
parser.add_argument('--sample_t_stride',
default=1,
type=int,
help='If larger than 1, input frames are subsampled with the stride.')
parser.add_argument('--train_crop',
default='random',
type=str,
help=('Spatial cropping method in training. '
'random is uniform. '
'corner is selection from 4 corners and 1 center. '
'(random | corner | center)'))
parser.add_argument("--scale_h", type=int, default=128,
help="Scale image height to")
parser.add_argument("--scale_w", type=int, default=171,
help="Scale image width to")
parser.add_argument('--train_crop_min_scale',
default=0.25,
type=float,
help='Min scale for random cropping in training')
parser.add_argument('--train_crop_min_ratio',
default=0.75,
type=float,
help='Min aspect ratio for random cropping in training')
parser.add_argument('--no_hflip',
action='store_true',
help='If true holizontal flipping is not performed.')
parser.add_argument('--colorjitter',
action='store_true',
help='If true colorjitter is performed.')
parser.add_argument('--train_t_crop',
default='random',
type=str,
help=('Temporal cropping method in training. '
'random is uniform. '
'(random | center)'))
parser.add_argument('--learning_rate',
default=0.1,
type=float,
help=('Initial learning rate'
'(divided by 10 while training by lr scheduler)'))
parser.add_argument('--momentum', default=0.9, type=float, help='Momentum')
parser.add_argument('--dampening',
default=0.0,
type=float,
help='dampening of SGD')
parser.add_argument('--weight_decay',
default=1e-3,
type=float,
help='Weight Decay')
parser.add_argument('--mean_dataset',
default='kinetics',
type=str,
help=('dataset for mean values of mean subtraction'
'(activitynet | kinetics | 0.5)'))
parser.add_argument('--no_mean_norm',
action='store_true',
help='If true, inputs are not normalized by mean.')
parser.add_argument('--no_std_norm',
action='store_true',
help='If true, inputs are not normalized by standard deviation.')
parser.add_argument('--value_scale',
default=1,
type=int,
help=
'If 1, range of inputs is [0-1]. If 255, range of inputs is [0-255].')
parser.add_argument('--nesterov',
action='store_true',
help='Nesterov momentum')
parser.add_argument('--optimizer',
default='sgd',
type=str,
help='Currently only support SGD')
parser.add_argument('--lr_scheduler',
default='multistep',
type=str,
help='Type of LR scheduler (multistep | plateau)')
parser.add_argument('--multistep_milestones',
default=[150, 225], # [150, 250] # default
type=int,
nargs='+',
help='Milestones of LR scheduler. See documentation of MultistepLR.')
parser.add_argument('--overwrite_milestones',
action='store_true',
help='If true, overwriting multistep_milestones when resuming training.')
parser.add_argument('--plateau_patience',
default=10,
type=int,
help='Patience of LR scheduler. See documentation of ReduceLROnPlateau.')
parser.add_argument('--batch_size',
default=128,
type=int,
help='Batch Size')
parser.add_argument('--inference_batch_size',
default=0,
type=int,
help='Batch Size for inference. 0 means this is the same as batch_size.')
parser.add_argument('--batchnorm_sync',
action='store_true',
help='If true, SyncBatchNorm is used instead of BatchNorm.')
parser.add_argument('--n_epochs',
default=200,
type=int,
help='Number of total epochs to run')
parser.add_argument('--n_val_samples',
default=3,
type=int,
help='Number of validation samples for each activity')
parser.add_argument('--resume_path',
default=None,
type=Path,
help='Save data (.pth) of previous training')
parser.add_argument('--no_train',
action='store_true',
help='If true, training is not performed.')
parser.add_argument('--no_val',
action='store_true',
help='If true, validation is not performed.')
parser.add_argument('--inference',
action='store_true',
help='If true, inference is performed.')
parser.add_argument('--inference_subset',
default='val',
type=str,
help='Used subset in inference (train | val | test)')
parser.add_argument('--inference_stride',
default=16,
type=int,
help='Stride of sliding window in inference.')
parser.add_argument('--inference_crop',
default='center',
type=str,
help=('Cropping method in inference. (center | nocrop)'
'When nocrop, fully convolutional inference is performed,'
'and mini-batch consists of clips of one video.'))
parser.add_argument('--inference_no_average',
action='store_true',
help='If true, outputs for segments in a video are not averaged.')
parser.add_argument('--no_cuda',
action='store_true',
help='If true, cuda is not used.')
parser.add_argument('--n_threads',
default=8,
type=int,
help='Number of threads for multi-thread loading')
parser.add_argument('--checkpoint',
default=10,
type=int,
help='Trained model is saved at every this epochs.')
parser.add_argument('--model',
default='r21d',
type=str,
help=
'(r21d)')
parser.add_argument('--model_depth',
default=18,
type=int,
help='Depth of resnet (18 | 34)')
parser.add_argument('--input_type',
default='rgb',
type=str,
help='(rgb | flow)')
parser.add_argument('--manual_seed',
default=1,
type=int,
help='Manually set random seed')
parser.add_argument('--accimage',
action='store_true',
help='If true, accimage is used to load images.')
parser.add_argument('--output_topk',
default=5,
type=int,
help='Top-k scores are saved in json file.')
parser.add_argument('--file_type',
default='jpg',
type=str,
help='(jpg | hdf5)')
parser.add_argument('--tensorboard',
action='store_true',
help='If true, output tensorboard log file.')
parser.add_argument('--distributed',
action='store_true',
help='Use multi-processing distributed training to launch '
'N processes per node, which has N GPUs.')
parser.add_argument('--dist_url',
default='tcp://127.0.0.1:23456',
type=str,
help='url used to set up distributed training')
parser.add_argument('--world_size',
default=-1,
type=int,
help='number of nodes for distributed training')
parser.add_argument('--val_freq',
default=1,
type=int,
help='evaluate on validation set every x epochs')
parser.add_argument('--loss_weight',
default=0.5,
type=float,
help='loss weight')
parser.add_argument('--use_audio',
action='store_true',
help='If true, load audio features to compute additional regularization term.')
parser.add_argument('--use_image',
action='store_true',
help='If true, load image features to compute additional regularization term.')
parser.add_argument('--normalization',
action='store_false',
help='normalization on features')
args = parser.parse_args()
return args